Robinhood Lets AI Agents Trade With Approvals Off. What That Setting Really Controls

Robinhood is putting AI agents inside its app that can research markets and place trades. By default, you approve each order. You can switch that off, and the agent can then place eligible trades on its own.

That makes Robinhood Agents a clean real-world example of a question every agent builder faces: what exactly does a human-approval toggle protect, and what does it leave open?

This is not investment advice, and I have not used the product. Everything below comes from Robinhood's support pages and press coverage as of October 5, 2026. Features are rolling out in phases, so check current details.

What did Robinhood announce?

At its HOOD Summit on September 29, Robinhood unveiled built-in agents, according to PYMNTS. It replaces the earlier May setup, where customers had to bring an outside agent and connect it through developer tools.

PYMNTS reports that more than 150,000 customers had opened agentic accounts by late September, and that agents use Robinhood's tools almost 30 million times a day, per unaudited figures from the company. A Frontier Models report says built-in agents are beginning to reach eligible customers in phases, and that a feature called Loops, for standing instructions that run in the background, is still forthcoming.

I read Robinhood's own trading with your agent support page for the controls. Press summaries differ on some details, so I treated the company page as the primary source.

What does the approval setting do?

Per Robinhood's support page, trade approvals are a safety control. With the setting on, the agent can propose trades, but you must review and manually place them in the app. With it off, the agent can place eligible trades without your confirmation of each order.

Two details matter:

• The default differs by route. Robinhood says approvals are on by default for built-in agents. Frontier Models, citing the same guidance, reports accounts connected through an external agent start with approvals off. Check which route you are on.
• Off does not mean unlimited. The page says certain trades may still require approval even when the setting is off. It does not list which ones, and Frontier Models notes the same gap.

In some states, the page says approvals must stay on for crypto trades.

What does the dedicated account protect?

According to the coverage, built-in agents trade from a separately funded account, so the agent can spend only the money placed there. The Modern Wallet's explainer describes it as leaving your primary portfolio untouched.

That is a real boundary on spending. Frontier Models adds a caveat worth keeping: Robinhood's documentation allows the agent to see portfolio information from other accounts, so separate funds do not make other holdings invisible. Trading authority and data access are separate permissions.

And the boundary does not protect the money inside it. Whatever you put in the agent's account can still be lost.

Who bears the loss?

Robinhood's own disclosure on the support page says: "You assume all risk for trades executed by AI agents and for any use of your data by third-party AI providers." It also warns that agentic trading involves significant risk, including the possible loss of the entire investment, and that strategies may be hard to monitor or stop in real time.

PYMNTS reports that Robinhood says it does not supervise, monitor, or audit the agents, and that the risk allocation applies whether approvals are on or off. Frontier Models reports the customer agreement treats authenticated agent instructions as final and binding when received. That is the company's contractual position, not a legal ruling.

The practical reading: the approval toggle is a customer-side control, and the customer carries the consequences either way.

Does pausing or revoking stop everything?

Here the sources disagree, which is worth knowing before you rely on a kill switch.

Frontier Models reports that pausing or disconnecting an agent does not itself reverse instructions already submitted or positions already taken, and that revoking access does not obligate Robinhood to cancel or unwind instructions already received. Robinhood's support page says that if you move to a restricted state, orders your agent placed earlier will not be canceled.

The Modern Wallet guide, by contrast, describes a one-tap kill switch that cancels active orders on revocation. That guide is about the earlier external-agent beta and is not Robinhood's own page, so I would not rely on it. I could not confirm that claim from the company's pages I read.

For any agent product, test what "stop" actually does before you give it money.

What can builders take from this?

This is my own checklist, not Robinhood's guidance.

1. Name what the approval gate covers. Is it every action, or only some? Robinhood's "certain trades may still require approval" shows why you should publish the exceptions.

2. Separate spend limits from data access. A dedicated balance limits spending. It does not limit what the agent can read.

3. Make the default safe, and make the opt-out explicit. Robinhood's built-in agents start with approvals on and ask the customer to confirm turning them off, per its setup page.

4. Say what stop means. Pausing new actions and undoing past ones are different. Document which you offer.

5. Make responsibility legible. A disclosure that moves all risk to the user is a legal choice. It does not replace controls that limit the damage.

6. Treat standing instructions as a new permission. Frontier Models notes that public materials do not say whether activating a Loop needs approvals off or a separate grant. Recurring authority deserves its own consent.

I wrote about the runtime side of this in the agent permissions article. A toggle in an app is a policy. A limit enforced in the system is a control.

FAQ

Can Robinhood's AI agents trade without my approval?

Yes, if you turn trade approvals off. Robinhood's support page says the agent can then place eligible trades without your confirmation of each order. Approvals are on by default for built-in agents, and some trades may still need approval.

Is the agent limited to the money in its own account?

According to the coverage, built-in agents trade from a dedicated funded account. Frontier Models notes the agent may still see portfolio information from other accounts.

Who is responsible if the agent makes a bad trade?

Robinhood's disclosure says you assume all risk for trades executed by AI agents. PYMNTS reports this applies whether approvals are on or off.

Does disconnecting the agent cancel its open orders?

Robinhood's page says earlier orders are not canceled if you move to a restricted state, and Frontier Models reports that revoking access does not unwind instructions already received. One third-party guide claims otherwise, but I could not confirm it from Robinhood's pages.

What are Loops?

Per Frontier Models and PYMNTS, Loops are standing instructions an agent runs repeatedly, such as a morning market check. They are described as coming soon, with no launch date.

Is this safe for retail investors?

I cannot say, and this is not investment advice. Robinhood itself warns of significant risk, including loss of the entire investment. Read the disclosures on its support pages before using any agent with real money.

What does this mean for people building agents outside finance?

The same questions apply anywhere an agent acts for a user: what needs approval, what is the agent allowed to see, what does stop actually undo, and who carries the loss.